Three quarters of advertisers told the IAB earlier this year their measurement approaches aren't working. Not one said their models fully represent all paid channels. Attribution, incrementality, marketing mix modeling: all underperforming on speed, accuracy, and trust.
The measurement industry's response was to make those same models run faster, stack them with incrementality testing and platform attribution, use AI to average out where they agree, and call it "unified truth." Unfortunately, the gap between what the models can see and what actually builds a business doesn't close when you accelerate a faster version of the same thing.
The most recent structurally ambitious versions of this fix arrived at Cannes last week. Mutinex and The Trade Desk debuted a system that runs MMM on four- and 12-week lookback windows refreshed monthly that then pipes those signals into programmatic buying. They call it outcome-based media activation and those testing the technology hailed it as an enabler for shifting from "reach" to "relevance" by tying every dollar to real business impact in something close to real time. The trade press called it the future of measurement without ever addressing what the model can't see.
Binet and Field spent decades analyzing nearly a thousand IPA campaigns to find out what actually drives profit growth. Specifically profit, not Return on Marketing Investment, not efficiency, but rather the number that compounds over time. The finding the industry spent the last decade selectively ignoring: short-term campaigns generate impressive spikes at low spend because they activate demand that already exists but investment in brand-building actually creates that demand.
Over-indexing on the first at the expense of the second boosts ROMI while steadily eroding profit growth. They're separate instruments measuring different things, and digital media made short-term measurement so clean, so fast, so efficient, so satisfyingly full of numbers, that the industry convinced itself it had found a single instrument that measured everything. We all bragged about finding a really good thermometer when in reality we needed a stethoscope.
A four-week MMM will show you retail media is working. Of course it's working. Lower-funnel performance media will show you what it's designed to show: people already primed, already in market, already with the brand in their head, already ready to convert. The broad reach TV campaign that put the brand in their head six months ago shows up in a four-week window as nothing, because the effect is diffuse, delayed, and spread across channels that don't close the loop in a biddable environment. The model recommends more retail media. The budget shifts. Next quarter the model recommends more retail media again. The cycle runs until the brand has hollowed itself out buying customers it used to attract for free, back when people knew who it was.
The acquisition economics of every industry that made this tradeoff are incredibly clear. DTC eCommerce CAC increased 40 to 60 percent from 2023 to 2025. Over eight years, the number is 222 percent. Platform inflation accounts for some of that but the bigger driver is structural: the industry is harvesting demand faster than it builds demand and measuring only the harvest. The brands with CAC 30 to 50 percent lower than competitors, and conversion rates 2.5 times higher, got there by doing brand-building work that never appears in a four-week performance report.
The AI bidding layer is making this even harder to reverse. Google’s PMax and Meta’s Advantage+ currently optimize aggressively for measurable signals: lower-funnel conversions, branded searches, retargeting. Demand creation doesn't register as a measurable signal on those timelines.
LiftLab framed the structural trap cleanly: a more precise model trained on biased budgets delivers a more accurate version of the same flawed answer. In short, visibility bias. Every efficiency gain from AI-optimized bidding compounds the same bias with the model getting better at telling you what it can see while what it can't see gets defunded quarter by quarter.
Which is why the Mutinex/Trade Desk Cannes announcement deserved harder scrutiny than it got. The channels where they demonstrated closed-loop optimization: search, open-web, and selected social. That list maps exactly onto the channels that perform well on short measurement windows. Brand-building channels are harder to connect to four-week outcome signals in a programmatic environment, so they don't make the demo.
When an FMCG company announces it's shifting from reach to relevance, someone should ask about the light buyers who account for most of the category's volume and who aren't actively thinking about the category or product until they're standing in the aisle. Reach is how you build the mental availability that makes future relevance possible.
A Harvard Business Review survey from May 2026 found that only 28% of organizations effectively convert measurement insights into action. Even more bewildering: companies with the most sophisticated measurement stacks have an execution gap as wide as the laggards.
Most of us built these tools to win budget arguments, not to understand how our businesses actually grow. Finance wants quarterly proof. Boards reward what's in the slide deck. CMOs fight for budget in rooms where a four-week ROAS chart beats a brand equity thesis before the sentence is finished. Marketing organizations want the glory of a Lion or Effie. It’s not the measurement systems fault though. They are very good at serving those masters.
The drift toward what's measurable on a short clock is organizational, and it's sustained by incentive structures that reward quarterly proof over durable growth. The harder question is whether anyone in your company has the standing to argue for a measurement window that matches your actual purchase cycle, and whether the people who control budget are willing to chase faster proof of lasting things instead of faster proof of faster things.
*** Views expressed on Push Pull Pivot are personal and shared in an individual capacity. They do not represent those of any current or former employer ***